Objective: Explore a specific application within a specific domain, identify three significant papers, and conduct a comparative analysis.

Steps

$1.$ Choose an Application Area: Choose any one application area from the list given below. You can choose your own domain also.

List of potential application area:

CO$2$ Emission Prediction

Cyclone prediction

Traffic Flow Prediction

Automatic music generation

Energy Consumption Prediction

Building Energy Optimization

Waste Composition Analysis

Predictive Air Quality Models

Application for cancer detection

Gender recognition using voice

Content Recommendation with Transformers

Medical Image Diagnosis

Speech Recognition

Speech Translation

Emotion Recognition in Social Media

Autonomous Navigation for Robots

Gesture Recognition for Human$-$Computer Interaction

Wildlife Classification

Real$-$time Language Translation

Human Activity Recognition from videos

Expression Recognition from images

$2.$ Identify Three Papers: Identify three significant journals which uses Deep Feedforward Neural Network $/$ CNN $/$ RNN $/$ Transformer networks $($any one has to be chosen$).$ You can use transfer learning for CNN also. Journal should be from reputed sources like IEEE$/$Springer or ACM that focus on the application of CNN$/$RNN$/$Transformer networks in your chosen domain. Upload all three PDFs as individual files on Canvas.

$3.$ Compare the architecture and methodologies used in the journals.

Create a Comparison Table: Compare the three papers and present your findings in a table with the following titles:

Group Number, member names, and BITS IDs

Domain

PAPER $1,$ PAPER $2,$ PAPER $3($with subheadings: Title, Authors, Year, Architecture of Deep Learning $($including the number of layers, types of layers, activation functions, and any unique features$).$ Network application $($e$.$g$.,$ feature engineering, classification, regression$),$ Training procedures $($e$.$ g$,$ training strategy, including optimization algorithms, learning rates, batch sizes, and regularization techniques$)$ Evaluation$/$Performance metric, Dataset used, URL if public dataset$)[*$Reference Comparison Table is given below$]$

Conclude: End the comparison with a proper conclusion highlighting your observations. Justify the choice of one paper over the others for implementation in Part B$.$

Submission: Upload the table and comparison as one PDF $($Filename: DomainName$\_$GroupNumber$).$

Expected Comparison Table $(5$ marks$)$

Title of the paper

Authors

Year of publication

Architecture of Deep Learning $($including the number of layers, types of layers, activation functions, and any unique features$)$

How is the network helping the overall task? eg: feature engg or classification or regression or all

Training procedures $($e$.$g$,$ training strategy, including optimization algorithms, learning rates, batch sizes, and regularization techniques$)$

Evaluation $/$ Performance metric used

Name of Dataset used. If a public dataset, provide the URL.

Conclusion: You must end the comparison with a proper conclusion highlighting your observations.

Implementation is deferred to assignment $2.$

Part B: Industry DL Product $(7$ marks$)$

Objective: Identify and understand any industry product and summarise the understanding

Instructions:

Identify DL Product: Identify any product that is used in any industry. Egs of products $--$ Product Recommendation in Amazon, Music recommendation in Spotify, ChatGPT, Dall$-$E$,$ Face tagging in Facebook, Bard $...$ Ensure that it is a product used int he industry. You can choose your own.

Identify the white paper associated with the above product, if any: Upload the white as PDF in the Canvas.

Summarise: In your own words, summarise the product.

Paragraph $1$: What is the objective of the product.

Paragraph $2$: What is the solution technology used. How is the solution achieving he objective mentioned earlier.

Paragraph $3$: What are the frameworks, algorithms, tools etc used for the developing the solution.

Paragraph $4$: What are the issues in the current solution.

Paragraph: Do you see any future scope in similar products.

Additional Instructions:

Journals can be chosen without any restrictions on impact factors or other indices.

Select three research papers within a single domain, each employing a different algorithm $($CNN$,$ RNN$,$ Transformers$)$ for comparative analysis.

Any ONE algorithm has to be used.

Dataset can be same or different.

For any queries, use the discussion forum.